Temperature-constrained cascade correlation networks(TCCCNs) were applied to the identification of the powder pharmaceutical samples of metronidazole based on near infrared(NIR) diffuse reflectance spectra. This work focused on the comparison of performances of the uni-output TCCCN(Uni-TCCCN) to multi-output TCCCN(Multi-TCCCN) by using near infrared diffuse reflectance spectra of metronidazole. The TCCCN models were verified with independent prediction samples by using the "cross-validation" method. The networks were used to discriminate qualified, un-qualified and counterfeit metronidazole pharmaceutical powders. The results showed that multiple outputs network generally worked better than the single output networks. With proper network parameters the pharmaceutical powders can be classified at a rate of 100% in this work. Also, the effects of neural network parameters including number of candidate nodes, type of transfer functions(linear, sigmoid functions and temperature-constrained sigmoid function, respectively) on classification were discussed.
The application of artificial neural network for pharmaceutical nondestructive quantitative analysis was investigated. The real data set from near infrared reflectance spectra of trimethoprim powder pharmaceutical were used to built up artificial network to predict unknown samples. The factors affecting network were discussed. A new network evaluation criterion, the degree of approximation, was employed. Owing to good nonlinear multivariate calibration nature of ANN, the predicted result was reliable. [WT5HZ]
